Yahoo DSP scales keyword expansion 600x via Amazon Bedrock integration
Yahoo has integrated Amazon Bedrock and Anthropic’s Claude 3.5 Sonnet into its Demand-Side Platform to automate and enhance search retargeting keyword expansion. The new generative AI-powered workflow reportedly improves keyword expansion rates by 600 times and audience reach by fivefold compared to the legacy Word2Vec-based system.
Key Takeaways
- Keyword expansion rates increased by 600x compared to the previous Word2Vec and locality-sensitive hashing architecture.
- Addressable audience reach grew fivefold, driven by the LLM's ability to capture semantic intent rather than just outdated phrase matches.
- Anthropic’s Claude 3.5 Sonnet v2 was selected following SageMaker Studio prototyping for its balance of expansion ratio and semantic similarity.
- A custom verification step uses embeddings to filter out low-similarity keywords, mitigating potential LLM hallucinations in production workflows.
Why It Matters
This architectural shift signals a transition from legacy machine learning to agentic, generative workflows within the programmatic stack. By moving to serverless infrastructure via Bedrock, Yahoo eliminates the overhead of model hosting while gaining the ability to swap foundation models as performance benchmarks evolve. For the broader ecosystem, it demonstrates how top-tier DSPs are weaponizing first-party search intent to defend market share against walled gardens as third-party cookies diminish. Watch for internal conversion metrics to clarify if this massive reach expansion maintains the high-intent signal strength traditionally associated with search retargeting.
Additional Context
The Amazon Bedrock integration aligns with Yahoo’s broader 'Yours, Mine, and Ours' AI strategy, which seeks to position its DSP as an interoperable backbone for the industry. In June 2026, Yahoo launched its 'Agent Network,' an open framework connecting advertisers with specialized AI agents from partners like DoubleVerify, LiveRamp, and Snowflake directly within the DSP. This approach allows advertisers to use their own proprietary models alongside Yahoo’s native tools, targeting a reduction in the fragmentation typical of multi-platform programmatic buying.
Simultaneously, Yahoo has prioritized trust and source transparency as it competes with Google and Amazon for search-driven ad spend. At its 2026 NewFronts presentation, the company introduced 'Yahoo Scout,' an AI-powered answer engine also built on Anthropic’s Claude. Scout serves as an 'intelligence layer' across the Yahoo portfolio, extracting deeper intent signals from Mail, Finance, and Sports to feed the DSP. Per MediaPost (May 2026), these signals provide a privacy-focused edge as advertisers look for cookieless alternatives.
The scale of this deployment reflects a maturing enterprise AI market. According to AWS financial data for Q2 2026, Bedrock customer spending increased 170% year-over-year as major organizations moved from experimentation to large-scale production. Amazon’s overall advertising services revenue grew 26% to $19.8 billion in the same period, highlighting the massive stakes for Yahoo as it seeks to scale its programmatic reach using the same underlying Bedrock infrastructure.
Read full article at aws.amazon.com
Enjoy our coverage?
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source